This paper introduces a multivariate kernel based forecasting tool for the prediction of variance-covariance matrices of stock returns. The method introduced allows for the incorporation of macroeconomic variables into the forecasting process of the matrix without resorting to a decomposition of the matrix. The model makes use of similarity forecasting techniques and it is demonstrated that several popular techniques can be thought as a subset of this approach. A forecasting experiment demonstrates the potential for the technique to improve the statistical accuracy of forecasts of variance-covariance matrices.
Becker, R., Clements, A., & O'Neill, R. (2018). A Multivariate Kernel Approach to Forecasting the Variance Covariance of Stock Market Returns. Econometrics, 6(1), 1-27. . https://doi.org/10.3390/econometrics6010007